Code Room
On-callHardoc-g572
Subject Ml train serve skewLevel Senior–Staff~40 minCommon in ML systems · Reliability & on-call interviewsIndustries Technology

Question

After a new pricing-model release, the model's online performance is much worse than its glowing offline eval: predicted prices are systematically biased low and the business sees margin erosion, though there are no errors. Investigating, you find the offline training pipeline computes 'normalized_demand' with a 7-day rolling window in a batch Spark job, while the online serving path computes the same feature in a separate Java service — and the two implementations disagree (the online one uses a different window boundary and unit). Walk through how you confirm the issue, mitigate, and prevent it.

What a strong answer looks like

Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.

Diagram & narrate the incident
Loading whiteboard…
Run or narrate your approach, then ask the coach.